AI Governance and the Policymaking Process: Key Considerations for Reducing AI Risk
Round 1
Reviewer 1 Report
The paper explores the policy options to reduce the risk of AGI on the public government level. Authors propose new subfield of AI governance where the focus is on policy-making strategies that consider AI. The main contribution is the policy-making cycle where authors present different frameworks for AI governance with possible problems and potential solutions, mainly with references to existing papers and books.
The paper is well structured, read easily and the English language used in the paper is appropriate - nonetheless, the paper should be proofread, because there are minor spelling and grammar mistakes.
My only issue with the paper is that the abstract is poorly written - the main point of the paper is at the end of the abstract. I suggest for the abstract to be rewritten.
Author Response
Note: We had some issues with our Word document being able to maintain its proper format (line numberings being moved around, weird spaces, etc), so we’re submitting the LaTex PDF as per Carolina’s suggestion. If you have any questions regarding the changes, feel free to ask. Also, we are re-submitting this as an essay
The paper is well structured, read easily and the English language used in the paper is appropriate - nonetheless, the paper should be proofread, because there are minor spelling and grammar mistakes.
We have corrected the spelling and grammar mistakes we have found.
My only issue with the paper is that the abstract is poorly written - the main point of the paper is at the end of the abstract. I suggest for the abstract to be rewritten.
We have rewritten the abstract to bring the main point of the paper to the beginning, and end with the conclusion of our paper.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Overall observation: good paper. Minor clarifications and corrections needed. Should be published
Please see the detailed review attached
Comments for author File: Comments.pdf
Author Response
Note: We had some issues with our Word document being able to maintain its proper format (line numberings being moved around, weird spaces, etc), so we’re submitting the LaTex PDF as per Carolina’s suggestion. If you have any questions regarding the changes, feel free to ask. Also, we are re-submitting this as an essay
Line 72: ”our ideal vision” or ”the ideal vision”? This “vision” is that of the authors or it describes a possible common vision etc.?
This was the words of the Allan Dafoe paper, not ours. We’ve added quotations to communicate this.
Line 80: suggestion: “a better/clearer launching point” instead of “a launching point”, due to the previous presentation of the authors on the works of other researchers on their theme here. Are the authors claiming their paper is a breakthrough?
Changed
Line 97: “mitigating” a risk is not annihilating the risk. How safe is “mitigating”only? Please clarify
We would never claim that risk can be completely rid of, especially when it comes how political systems interact with risky behaviors and technologies. We’re unclear how discussing this distinction would contribute to our paper.
Lines 166-189 = clear enough. See the previous observation on the Line 80
See previous response
Lines 190-198 = good. Is there a newer model that the Lowi’s one?
There are critical reconceptions of Lowi’s model, but it seems that Lowi’s model is still heavily cited and used. Exploring other ways of building typologies of AI policies should be reserved for future research
Lines 215-215 = typo correction needed for –
Corrected
Lines 231-232 = please rephrase
Corrected
Line 245: suggestion: “surprisingly” instead of “incredibly”, if the authors agree
Because other authors have examined the framing effect in regards to AI risk, saying that it’s surprising would not be appropriate.
Lines 344-346 = very good Suggestion: could be the Line 329 partially rephrased as “Policy Windows of Time”?
This is definitely clever and we like it, but think that combining terminology would be confusing for the reader, especially if they want to look further into the political science literature.
Line 367: Suggestions: “require/strongly require” instead of “begs” and “appropriate” instead of “serious”
Changed
Line 421 = typo correction needed –
Changed
Line 483: Suggestion: instead “one step”, may be “a key step”
Changed
Lines 513-523 = clear enough. However, I would suggest adding a short clearer statement on the “complexity”, accordingly to the non-linearity and non-predictability of the complex systems behavior
Added “These problems are non-linear, very hard to predict, and may have the traits of wicked problems in the sense that solving one problem can create new problems.”
Lines 534-539 = enough clear. However, it should be accommodated with the “complexity”. Question: it is about complexity or complicatedness, only?
This has been addressed in the previous point.
Lines 571-572: Observation: seems the phrase is incomplete. Please correct/add the required words after “that”
Corrected
Line 594: Suggestion: please replace “begs” as suggested above at Line 367
Corrected
Line 601: Suggestion: It seems the authors should introduce an “and” between “broad overview” and “to be an”
Corrected
Lines 601-604 = good OBSERVATION: there is a problem with the lines counting after Line 604
Lines 605-612 = very good. OBSERVATION: there is a problem with the lines counting after Line 612 Lines 613-617 = excellent OBSERVATION: there is a problem with the lines counting after Line 617
There was a problem with converting the Latex PDF to a word file. This should be corrected.
Lines 680-956 Suggestion: “Ibidem” instead of “Ibid”
Corrected
Author Response File: Author Response.pdf
Reviewer 3 Report
This is a well-written report on a very relevant/current/emerging topic but it’s not quite clear, as a research paper, what the contribution is.
Although the background discussion is sound and (probably) correctly highlights limitations with current discussion regarding AI policymaking, the subsequent attempt to deal with it lacks focus. Specifically, although the argument for a new approach is well-made, it doesn’t result in much. Section 4 only really contributes a discussion of areas into which a better analysis *might* be subdivided. It’s more of a top-level taxonomy than any sort of framework and neither is there any attempt to independently justify it.
And this is precisely where the paper could/should be improved. The discussion needs to be turned into a more formal framework and there needs to be proper evaluation of it. This might perhaps take the form of independent expert assessment or application to known case studies … or something else. But there needs to be *some* analysis and evaluation – otherwise there’s nothing firm to base any conclusions upon.
If we can get it to the point where the content is acceptable then there are a few oddities of English to look at: e.g. L85/93 ‘do/doing … overview’ (yuk!), e.g. L116 inconsistent use/avoidance of apostrophe abbreviations, L120 mismatched noun/verb/plurals, e.g. L162 Not normal to use both (year)[ref], L175 -> ‘useful overall’, etc.
Author Response
Note: We had some issues with our Word document being able to maintain its proper format (line numberings being moved around, weird spaces, etc), so we’re submitting the LaTex PDF as per Carolina’s suggestion. If you have any questions regarding the changes, feel free to ask. Also, we are re-submitting this as an essay
We are re-submitting this as an essay, because we believe this better fits the idea of what we’re going for. We are not submitting this as an empirical finding, but rather the starting point of a discussion that could lead to empirical research.
This is a well-written report on a very relevant/current/emerging topic but it’s not quite clear, as a research paper, what the contribution is.
This paper, as another editor neatly summarized: “...explores the policy options to reduce the risk of AGI on the public government level. Authors propose new subfield of AI governance where the focus is on policy-making strategies that consider AI. The main contribution is the policy-making cycle where authors present different frameworks for AI governance with possible problems and potential solutions, mainly with references to existing papers and books.” We are hoping the main contribution that this paper has is to get other researchers to think about this problem from this perspective and progress the line of research.
Section 4 only really contributes a discussion of areas into which a better analysis *might* be subdivided. It’s more of a top-level taxonomy than any sort of framework and neither is there any attempt to independently justify it.
Yes, this is a top-level analysis of the problem. The purpose of this paper is to open up further questions of exploration. From what we have gathered, nobody has explored the area of AI governance from the perspective of policymaking strategies. We cover some of the main literature that talks about the policymaking process, and how it might apply to AI governance. There are many entry points of discussion (i.e. framing effects, how to get policies on formal agendas, how policies should be implemented, etc) and each area would likely develop their own frameworks and would complicate this paper. We believe that having this high level framework will let researchers have a better understanding of the issue.
And this is precisely where the paper could/should be improved. The discussion needs to be turned into a more formal framework and there needs to be proper evaluation of it. This might perhaps take the form of independent expert assessment or application to known case studies … or something else. But there needs to be *some* analysis and evaluation – otherwise there’s nothing firm to base any conclusions upon.
See our previous response. We are publishing this as an essay, not empirical findings, so this is outside the scope of our paper.
If we can get it to the point where the content is acceptable then there are a few oddities of English to look at: e.g. L85/93 ‘do/doing … overview’ (yuk!), e.g. L116 inconsistent use/avoidance of apostrophe abbreviations, L120 mismatched noun/verb/plurals, e.g. L162 Not normal to use both (year)[ref], L175 -> ‘useful overall’, etc.
Done
Author Response File: Author Response.pdf
Reviewer 4 Report
The aims are clearly described in the Introduction to the text: "The goal of this paper is to attempt to create a launching point for discussions on the key considerations of the policymaking process for AI governance and the political considerations underpinning policy solutions for AI risk". These aims are considered to be relevant, carrying out in general an in-depth work and with precision.
The main weakness found in the work is observed in the methodology. The criteria followed are not evident, especially when it comes to selecting some references and not others, one context and not another.
The following comments and suggestions are specifically made:
(1) It is necessary to describe methodology followed in the research, describing the motivation of the identified case studies and the selected references, sources of knowledge, etc. This section could be fed with information provided in section "7. Appendix/Summary", and an assessment should be made as to whether section 7 itself is necessary. Due to the nature of the research, it may not be necessary for there to be a section with the name "Methodology", but it must be clearly described.
(2) The insertion and format of references throughout the text should be reviewed.
(3) Line 67. It is necessary to give a reference to "Future of Humanity Institute's research".
(4) Check line 174.
(5) Certain details of the paragraph, such as line 494, would fit more properly into the Introduction.
(6) Line 503. Check in relation to the reference.
(7) Line 572. The sentence is not finished.
(8) Paragraph 7. Review whether this paragraph is necessary after including the Methodology.
Author Response
Note: We had some issues with our Word document being able to maintain its proper format (line numberings being moved around, weird spaces, etc), so we’re submitting the LaTex PDF as per Carolina’s suggestion. If you have any questions regarding the changes, feel free to ask. Also, we are re-submitting this as an essay
(1) It is necessary to describe methodology followed in the research, describing the motivation of the identified case studies and the selected references, sources of knowledge, etc. This section could be fed with information provided in section "7. Appendix/Summary", and an assessment should be made as to whether section 7 itself is necessary. Due to the nature of the research, it may not be necessary for there to be a section with the name "Methodology", but it must be clearly described.
We are re-submitting this as an essay to more accurately reflect the epistemological standing of our paper. We have clarified that the papers cited are the primary pieces of literature for their respective theories. However, because this is a more exploratory piece, and not empirical, we do not believe we should have a specific methodology.
(2) The insertion and format of references throughout the text should be reviewed.
Corrected
(3) Line 67. It is necessary to give a reference to "Future of Humanity Institute's research".
Corrected
(4) Check line 174.
Corrected
(5) Certain details of the paragraph, such as line 494, would fit more properly into the Introduction.
The amount of money being spent on AI safety research is in reference to the concepts of local capacity and will in terms of policy implementation, and thus belong to the to this part of the discussion. We would argue that it does not add to the conversation in the introduction, which primarily deals with historical discussion on AI policy and not technical research.
(6) Line 503. Check in relation to the reference.
Corrected
(7) Line 572. The sentence is not finished.
Corrected
(8) Paragraph 7. Review whether this paragraph is necessary after including the Methodology.
We believe it’s necessary given that we’ve changed from article to essay.
Author Response File: Author Response.pdf
Reviewer 5 Report
AI Governance and the Policymaking Process: Key Considerations for Reducing AI Risk
This paper identifies a preceding, meta-level problem of how the space of possible policies is affected by the policymaking process itself.
Interesting discussion, but i will recomend to resubmit as an "essay".
Use impersonal form ( remove "we" and use "this paper")
Line 5 (abstract). remove (1), (2), etc... just mention them.
"(1) problem identification/agenda setting, (2) policy formulation, (3) policy adoption, (4) implementation, and (5) evaluation."
to problem identification/agenda setting, policy formulation, policy adoption, implementation, and evaluation.
Line 15 "humanity. [1]" ---> "humanity [1]." Check all references, there are a lot of them in the paper.
Check references format.
Reference 51 ????
Reference 57, 58, 59 ... What means Ibid
Author Response
Note: We had some issues with our Word document being able to maintain its proper format (line numberings being moved around, weird spaces, etc), so we’re submitting the LaTex PDF as per Carolina’s suggestion. If you have any questions regarding the changes, feel free to ask. Also, we are re-submitting this as an essay
Interesting discussion, but i will recommend to resubmit as an "essay".
Thank you for the suggestion! We have done so.
Use impersonal form ( remove "we" and use "this paper")
Done
Line 5 (abstract). remove (1), (2), etc... just mention them.
Done
"(1) problem identification/agenda setting, (2) policy formulation, (3) policy adoption, (4) implementation, and (5) evaluation." to problem identification/agenda setting, policy formulation, policy adoption, implementation, and evaluation.
Done
Line 15 "humanity. [1]" ---> "humanity [1]." Check all references, there are a lot of them in the paper.
Done
Check references format.
Done
Reference 51 ???? Reference 57, 58, 59 ... What means Ibid
Ibid (or Ibidem) means it’s from same source that was mentioned in the previous reference.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
OK to accept as an essay.
Reviewer 5 Report
All proposed changes in previous round have been considered. Now the paper has been significantly improved.
